The controller spawns worker goroutines/threads. Each worker:

LANBench operates using a Client-Server model.

This is your LLM inference server exposed via a REST API or gRPC endpoint. LANBench requires standard request/response formats (typically /v1/completions or /v1/chat/completions).

Many engineers place NGINX or Cloudflare Tunnel in front of their LLM. Run LANBench directly to the LLM server, then run it again via the proxy. If token speed drops by 30%, you know your proxy configuration is the bottleneck.

LANBench is a lightweight, portable network benchmarking utility designed specifically for Local Area Networks (LANs). It is used to measure the throughput (speed) and stability of network connections between two computers. Unlike internet speed tests that measure connectivity to an outside server, LANBench focuses purely on the performance of the internal network infrastructure—switches, routers, cables, and Network Interface Cards (NICs).

It is particularly popular among system administrators and network engineers because it requires no installation, is free for personal and commercial use, and provides a simple graphical interface for generating network load.